The COALA-2020 campaign took place at Cataract Scout camp (34°14′44′′ S, 150°49′26′′ E) located 20 km north-northwest of Wollongong on the east coast of New South Wales (NSW), Australia. The site is surrounded by a heavily forested area mainly stocked with Eucalyptus species (see Fig. 1). North of the sampling site is a four-lane arterial road connecting the M1 motorway on the east coast with southwestern Sydney. Other possible anthropogenic sources impacting the site are two underground coal mine heads, located 1.5 km to the northeast and 2.5 km to the north. Further afield sources include the Sydney suburban area (around 18 km northwest), Sydney city (45 km north), Wollongong urban area, and Port Kembla steelworks in the southern part of Wollongong (28 km southeast).

The campaign was conducted from 17 January to 23 March 2020. The first period of the campaign (17 January to 5 February) was heavily impacted by smoke pollution from the bushfires affecting the region. On 5 February, a substantial rain event extinguished the fires and cleared the atmosphere of residual smoke pollution (Mouat et al., 2022; Simmons et al., 2022). The smoke pollution period has been removed from the analysis presented here because we focus on understanding atmospheric processes during more normal conditions. Thus, this paper presents the analysis of BVOCs alongside anthropogenic emissions and their role in NPF during the second part of the COALA-2020 ambient measurements campaign running from 5 February until 17 March 2020.

The instruments deployed in the campaign are presented in Table 1. They included an air quality monitoring station owned and operated by the NSW Government Department of Climate Change, Energy, the Environment, and Water (DCCEEW), located approximately 10 m away from the main sampling line for VOCs. This station included measurements of temperature, wind speed and direction, PM₁₀, PM_2.5, O₃, SO₂, NO_x, CO, and visibility. Inlet heights on this station were between 4.5 and 5.6 m above ground level. All NSW air quality monitoring stations are accredited by the National Association of Testing Authorities (Australia); however, it should be noted that these instruments are targeted at regulatory standards and are not research grade. In particular, this means that measurements made close to the detection limits are likely to be inaccurate and should be interpreted as indicative measures rather than accurate quantitative measures of atmospheric concentrations.

VOCs were measured using a proton transfer reaction mass spectrometer (Ionicon PTR-TOF-MS 4000), which operated with a mass range spanning m/z=18–256. The drift tube was held at a temperature of 70 °C, pressure at 2.60 mbar, and an electric field to molecular number density ratio of 120 Td. The instrument was housed in a separate climate-controlled unit. Samples were drawn from an inlet on a 10 m mast through a 20 m long PTFE line using a bypass flow of 1.2–3 L min⁻¹. Calibrations were made on site using standardized cylinders containing 17 compounds, including isoprene, monoterpenes, methyl vinyl ketone (MVK) & methacrolein (MACR), benzene, C₈-aromatics, and C₉-benzenes (Mouat et al., 2022). Mass spectra were integrated to produce data at 1 min temporal resolution. Mole fractions were further averaged on a 5 min basis.

A suite of aerosol instruments were operated within the Atmospheric Integrated Research facility for Boundaries and Oxidative eXperiment (AIRBOX) container (Chen et al., 2019). Sample air was drawn from a common aerosol bypass inlet. The inlet was located 5 m above ground level for the following instruments:

An Ultrafine Condensation Particle Counter (UCPC TSI 3776) was used to measure condensation nuclei number concentrations greater than 3 nm (CN₃) (TSI Incorporated, Shoreview, MI, USA). The instrument was operated at a sample flow rate of 300 mL min⁻¹. Measurements were recorded at 1 Hz temporal resolution.

The method proposed by Dal Maso et al. (2005) was used to classify the particle size distribution data. To apply the method, the particle number density plots were made for each day during the campaign, and the plots were visually inspected to identify if an event occurred on that day. A day of data was classified as an event if there was nucleation and growth up to 25 nm for at least 2 h.

Once the events were classified, a logarithmic fit was applied to determine the geometric diameter of each mode. The data were manually divided into chunks of 10 min to visually inspect and determine the number of modes and the geometrical diameter range of each event (nucleation <25 nm, Aitken 25–100 nm, accumulation >100 nm). Once those parameters were defined and included in the code, each event was divided into periods of time with similar distribution modes.

For illustration, a hypothetical event lasting 2 h was divided in two: 1 h with two simultaneous particle modes (nucleation and Aitken) and then 1 h with just one particle mode (Aitken). This was done to estimate an accurate geometrical particle diameter based on the number of modes. This avoided the problems of changes in the number of modes in the sample. Finally, the data were merged again to obtain a time series of the number of particles predicted with the fit, number of modes predicted, and geometrical particle diameter.

The algorithm works by providing the number of modes observed in the input dataset. Then, it selects the provided model equation for each mode number and iterates over 100 fits looking for the best fit. The Bayesian information criterion (BIC) was used to identify the best fit, looking for the lowest values. Once the best fit was selected, the total particle number estimated by the model was compared with the sample record for each sample to assure it was within a 5 % difference compared to the total particle number reported in the sample. The result was then visually checked to find the geometrical diameter and how it compares to the distribution size plots from the raw aerosol distribution size data. Once the model was considered representative and accurate enough, the growth rate for each event was determined using a simple linear regression of the change in the geometrical diameter in time from nucleation to Aitken and eventually to the accumulation mode.

Of the 40 d included in the analysis, nine did not have any data. Of the 31 d with data, 12 (39 %) showed clear NPF events, nine (29 %) were considered undefined, and 10 (32 %) did not have enough data or were classified as a non-event The percentage of days with NPF is similar to those of other sites in forested areas in the Northern Hemisphere (Kalkavouras et al., 2020; Uusitalo et al., 2021). Thirty-nine percent of days with NPF events and 29 % with undefined events implies a highly reactive atmosphere even in this rural area, with some anthropogenic influence of mobile sources and occasionally a coal-fired power plant in the Hunter Valley region.

Figure 2 illustrates the time series of an NPF event observed on 11 February 2020. The NPF event commenced at 8 a.m., preceded by a peak in both SO₂ concentrations and the estimated H₂SO₄ proxy. The shaded area in the plot highlights the growth period, which is marked by an increase in mode diameter and condensation sink. Ozone is also increasing at this time. The increase in aerosol SO42- and organics during this period shows the influence of this reaction chemistry on particles larger than 100 nm in the aerosol size distribution. We estimated the H₂SO₄ proxy using the rural model developed by Dada et al. (2020). This model was chosen from among the options because the environmental conditions under which it was derived are the most similar to those of our sampling site. The equation used to estimate the H₂SO₄ proxy was H2SO4rural=-CS2×2×10-9+CS2×2×10-92+SO22×10-9×9×10-9×GlobRad12, where CS is the condensation sink, SO₂ is the concentration of SO₂, and GlobRad is the global radiation obtained from the photosynthetic photon flux density (PPFD) values, with GlobRad = 0.327*PPFD.

Of the 12 d with NPF, four occurred during the night or early morning (before sunrise) and eight during the day. The time series of SO₂, NO_x, ozone, VOCs, and the aerosol composition were used to identify which variables triggered and influenced the aerosol formation and growth. Of the 12 event days of NPF, six include VOC data and eight include aerosol composition data, noting that the composition data are not applicable to particles <100 nm and only three events led to accumulation-sized particles (diameter >100 nm). The data available for each event are summarized in Table 2.

From the daily time series of all available variables over the 12 d of NPF events, it is evident that SO₂ frequently triggers or at least influences the particle formation. However, the trigger for nighttime events seems to be NO₂-related chemistry, but without complementary measurements, it's unclear. To group the common factors influencing NPF for daytime and nighttime events, a comparison of the growth rate was used to determine whether the rates were similar during the day and during the night.

The estimated growth rate is presented in Fig. 3. Only four of the nine events during daytime (upper panel of Fig. 3) had a representative Pearson coefficient (R>0.6); the remaining five events did not have a stable linear growth and are not shown in the plot. The events that showed unstable growth patterns suggest a highly variable condensation source, possibly resulting from changing H₂SO₄ concentrations. This is complicated further by changing wind directions.

Some events highlight how the dynamic nature of daytime concentrations of O₂, NO₂, and O₃ complicate the analysis (see Fig. S1 in the Supplement). Nonetheless, these events provide insight into the factors that may drive the growth and particle formation and hence were included in all the analysis. Event 15-02-2020 in Fig. 3 is an example of how the geometric particle diameter can change when there is rapid growth. The first part of the regression shows a slower growth rate. After the sixth hour of slow growth, the rate increases substantially, attributed to an increase in H₂SO₄ around this time. Shortly after this accelerated growth, there is a wind change from northerly to southerly (Fig. S4). Following the southerly wind shift, a lower condensation sink and higher relative humidity likely contributed to the geometric particle diameter (Gdp) increase via enhanced condensation and water uptake. Declining tracer levels SO₂ and NO_x indicate that local particle growth mechanisms were likely dominant over the influence of a new air mass up to the seventh hour, when increases in NO_x and SO₂ are observed.

In contrast to the daytime events, all the nighttime events were stable enough to determine the event growth rate. The growth rate varied considerably between events (see lower panel of Fig. 3) and most likely reflects differences in the factors driving the particle formation between these episodes. The specific oxidation pathways that were active during each event likely had a direct impact on the observed differences in growth rates. These reaction pathways might include monoterpene ozonolysis and condensation over previously formed clusters (Liu et al., 2023; Wang et al., 2023) or oxygenated VOCs (OVOCs) brought to the site and condensed on formed seeds, possibly initiating nucleation (Bianchi et al., 2019; Higgins et al., 2022). Some of these processes were observed during the campaign and will be further explored in the nighttime events section.

From the time series analysis of all daytime events (see Figs. 2, 5–6, and S1–S4), four key points were identified for NPF in the area:

SO₂ arriving at the site appears to trigger nucleation and growth events.

On some occasions, the SO₂ plume might last for a couple hours, as shown in the first part of the event on 11 February 2020 (see Fig. 2), while at other times, there are multiple peaks of high SO₂ measured at the site, as shown in several other events in the record (e.g. Figs. S2, S3, S4). However, subsequent nucleation was observed on every occasion in which SO₂ was observed above the detection limit at the site, and growth occurred within 0 to 150 min after the SO₂ was first detected. The time window difference between events reflects the influence of conditions at the start of a particle growth event. To highlight this phenomenon, a cross correlation between SO₂ and the aerosol mass of aerosol SO42- time series obtained from the TOF-ACSM and the measured particle number concentration (CN₃) was applied. Figure 4 shows the Pearson correlation between SO₂ and the CN₃ and aerosol SO42- in a window period of 4 h, i.e. starting 2 h before the nucleation commenced and ending after the first 2 h of the event. This time window captures the SO₂ influence on the particle formation. Each line/point shows the correlations at 0, 30, 60, 90, 120, and 150 min lagged for each daytime event. The dotted blue lines show where the lagged correlation is significant at (r>0.5).

To interpret Fig. 4, we can use the event on 11 February (black line) as an example. Here, the correlation between SO₂ and CN₃ becomes significant (at r> 0.5) if the SO₂ time series is lagged 120 min with respect to the aerosol data, and the correlation between SO₂ and aerosol SO42- becomes significant after 3 h. This means that if we move the SO₂ time series 2 h forward, it will be better correlated with the particle number concentration, accounting for the reaction time of SO₂ to produce H₂SO₄ and enhance/trigger the particle formation under the conditions in the atmosphere at the time. Usually, the SO₂ correlation with aerosol SO42- needs a longer lag time to be significant. This is a potential indication of the order in which the chemical reactions happen. First, we will see oxidation of the SO₂ to H₂SO₄, then nucleation, and finally growth in mass when there is condensation or coagulation near CCN sizes. Using time series analysis as shown here can provide more evidence when the chemical mechanisms are known but observations of other variables are not available.

A similar result is observed for other events at different lagged times. The difference in the time necessary to achieve a significant correlation between SO₂ and the particle number seems to be related to the quantity of VOCs available when the SO₂ plume arrives at the site. This aligns with our understanding of the transition from nucleation to particle growth. In the early hours, observed monoterpene levels are sufficient to drive nucleation through ozonolysis and subsequent HOM formation (Iyer et al., 2021; Kirkby et al., 2023; Wang et al., 2023). Particle growth was observed later in the day (see Fig. 2 for example), likely driven by the condensation of OVOCs. The increase in the sulfate fraction observed in the ACSM supports the condensation of sulfate-related species onto the growing particles. Events on 15 February and 6 and 7 March showed the highest correlations within the first 30 min of lagging the data. Common to these events were relatively high levels of monoterpenes (∼1 ppb either directly observed or inferred from high PAR and temperature) in the hour before NPF detection at the site (see Figs. S1, S2, and S4). The elevated monoterpene levels and subsequent ozonolysis likely initiated particle formation during these times, with the available H₂SO₄ further facilitating nucleation. The HOM proxy (monoterpenes*ozone; e.g. Zhang et al., 2024) also peaked during this period, supporting the idea that HOM formation via ozonolysis was a dominant oxidation pathway driving initial nucleation.

The event on 8 March also met this condition (see Fig. 5), although it exhibited a relatively low growth rate. Elevated isoprene and MACR + MVK concentrations during this event suggest the potential for isoprene to suppress new particle formation, as described by Heinritzi et al. (2020). Higher isoprene levels after 12:00, accompanied by increased MACR + MVK, coincided with a decline in the number of smaller particles (although the CN₃ data are incomplete). This is the first step in the reaction chain to produce C₁₅ dimers. This observation aligns with the HOM proxy (monoterpenes × ozone): higher proxy values corresponded to periods of higher particle numbers, while a decrease in the HOM proxy coincided with a decrease in particle numbers and an increase in MACR + MVK products, suggesting a shift towards more isoprene-influenced atmospheric chemistry. Concurrently, increases in the organic and sulfate fractions, along with the condensation sink, indicate a shift towards conditions favouring the growth of existing larger particles through condensation and coagulation, rather than nucleation events.

The 11 and 16 February events had similar arrival times for the SO₂ pollution (08:00 to 09:00), although the photochemistry was not fully active yet (see H₂SO₄); monoterpenes levels were consistently high during all the campaign (∼0.4 ppb based on the days with data), enough to promote nucleation. This presumption is supported by looking at the event on 16 February (see Fig. S3). In this event, a first peak of SO₂ at 08:00 started nucleation, but then condensation or coagulation dominated, favouring growth. The CN₃–CN₁₄ data show that after that initial nucleation period, the particle number was dominated by the >14 nm fraction. Multiple SO₂ plumes reaching the site produced higher ratios of H₂SO₄ but promoted growth to larger particles sizes, particularly on the sulfates fraction that correlates with the SO₂ peaks. In the evening, there were a couple of small particle bursts that were quickly coagulated on larger-size particles.

On 10 March (see Fig. 6), a sharp decline in high monoterpene concentrations was observed just before the aerosol event. The aerosol growth phase was then observed to correlate with peaks in SO₂ and NO_x, as well as elevated levels of isoprene. This suggests monoterpene ozonolysis initiated nucleation, and the observed particle growth coincided with periods indicative of increased atmospheric pollution, potentially contributing condensable material.

For all daytime events, SO₂ and NO₂ are significantly correlated, with a Pearson correlation of 0.78, suggesting a common source for both pollutants. The closest source of combustion products is the Appin Road located north of the sampling site. Given that the sampling site is away from other possible sources of SO₂ and NO₂ and that wind speeds during most of the campaign were relatively low (see Fig. S6), combustion from mobile sources is considered the most likely source of both compounds, but there might be some influence of more distant coal-fired power stations. Another factor to contribute to this theory is that the SO₂ levels were higher during the day, when most of the commuting takes place, i.e. when there is higher vehicle density on the roads. The intermittent SO₂ and NO₂ peaks suggest the influence of mobile sources with poor emission control onboard. The effects of vehicles with poor emission control technologies on ambient concentrations of SO₂, NO_x, anthropogenic VOCs, and PM have been seen in different studies (Kari et al., 2019; Phillips et al., 2019; Smit et al., 2019), and the legislation controlling fuel standards and emissions is relatively lax in New South Wales (Paton-Walsh et al., 2019).

During the COALA-2020 campaign, many events, such as the one on 16 February (Fig. S3), exhibited elevated gas-phase SO₂. The availability of monoterpene to form highly condensable ULVOC / ELVOC is crucial in the observed events. While the oxidation products of isoprene can also condense on pre-existing particles (Stangl et al., 2019), the dominant pathways and their efficiency are likely driven by monoterpenes. Although VOC data were not available for 16 February, the consistent diurnal profile of VOCs observed throughout the remaining dataset (Fig. S5) suggests enhanced monoterpene and isoprene availability during the daytime. Under these conditions of available BVOCs, particle growth was frequently observed, suggesting a contribution from condensed organic material. As the night approaches and BVOC emissions decrease with temperature, the remaining OVOCs can undergo further oxidation, forming less volatile species that are more prone to condensation on existing particles. However, the limited availability of VOCs after their consumption (estimated around 22:00 based on diurnal cycles in Fig. S5) likely limits further growth.

When there is negligible SO₂ in the atmosphere but high VOC concentrations (particularly monoterpenes), auto-oxidation processes can be initiated, potentially leading to both nucleation and subsequent particle growth (Bianchi et al., 2019). Growth was observed during the first event on 10 February (see daytime data in Fig. 7) despite low SO₂ and may be related to the condensation of HOMs formed through monoterpene auto-oxidation. The average concentration of monoterpenes during the campaign in the morning was often sufficient to initiate reactions leading to ULVOC that favour both new particle formation and the growth of pre-existing particles.

Australia experiences an isoprene-dominated atmosphere (Emmerson et al., 2016; Ramirez-Gamboa et al., 2021), and the chemical balance in the atmosphere can rapidly change, particularly in the hotter seasons when more isoprene is emitted. While SOA formation on pre-existing particles can involve molecules with relatively high saturation vapour pressures, new particle formation critically depends on molecules with extremely low saturation vapour pressures due to the Kelvin effect (Tröstl et al., 2016). Heinritzi et al. (2020) showed that reducing C₂₀ formation (α-pinene oxidation in the presence of isoprene) to favour C₁₅ formation reduces nucleation rates. However, it is also important to highlight that C₁₅, C₁₀, and even C₅ oxidation products from isoprene oxidation can contribute to SOA mass on existing particles. Therefore, in Australia's isoprene-dominated environment, higher isoprene to monoterpene ratios could lead to a greater production of C₅ and C₁₅ products that contribute to particle growth on existing aerosols (and SOA mass) while simultaneously hindering new particle formation by reducing the formation of C₂₀ dimers from monoterpenes.

We observed three nighttime events during COALA. Unfortunately, none of these events coincided with all datasets being collected, which limits our ability to discuss the reactions driving the nighttime events. Consistent between all nighttime events is an increase in particles (CN₃), elevated NO₂, and an increasing condensation sink. Unfortunately, the NO_x instrument available in this study was not ideal for this type of measurement for several reasons: it is not designed to be sensitive to the low NO_x levels observed in rural areas; it is not capable of separating NO_x from NO_y; and it was set up to calibrate in the night hours between 01:00 and 02:00 every day. Nonetheless, during the nighttime events, the particle size distribution data and the CN₃ data showed particle formation and growth from the nucleation to Aitken modes when there were considerable increases in NO₂ and simultaneous decreases in ozone.

When VOC data were available, monoterpene concentrations were moderate and increased steadily during the event (5 February and 9 March). Isoprene was high at the start of the event on 5 February (see Fig. 8); however, the sudden decrease in isoprene concentration likely coincides with sunset on that day. When aerosol composition data were available (10 February), aerosol organic, nitrate, and sulfate concentrations increased during the event. When ozone data were available, concentrations decreased slightly during the course of the event.

The frequency of nocturnal events observed in this study is lower than observed previously at a nearby location (Tumbarumba, a eucalypt forest site located 300 km southeast of Cataract; Suni et al., 2009), where, in the summer of 2006, nocturnal NPF events were observed on 32 % of the analysed nights and occurred 2.5 times more frequently than daytime events. Simulating the NPF at Tumbarumba, Ortega et al. (2012) was able to reproduce the observations from Tumbarumba by ozonolysis of 13-carene to initiate nucleation and a-pinene to grow particle diameters. Ozonolysis of limonene was found to contribute to both nucleation and aerosol growth. The lower frequency observed in our study may be linked to the apparent inhibition of nucleation by NO₂, which nocturnally can react with O₃ to form nitrate radicals. Li et al. (2024) suggest that even trace amounts of NO₃ radicals suppress the NPF.

Figure 9 shows the mass fraction of the PM₁ aerosol mass measured in the ACSM. Most of the daytime events show a similar mass fraction distribution. The organic fraction is the largest mass fraction, followed by sulfates, ammonium, nitrates, and chlorides. We observed higher sulfate mass fractions in days with higher SO₂ availability such as the events on 16 February and 8 March, where the average sulfate mass fraction was larger or similar to the organic fraction (see Fig. 9). These two events also display the highest proportion of ammonium during daytime events. The overall mass during nighttime is much lower than during daytime, likely related to the lower concentrations of VOCs available during the night, resulting in growth not reaching sizes where it was detectable by the ACSM. Even with less total mass during the night, the contribution of each fraction is similar to the daytime events. The most notable difference between the mass fractions during daytime and nighttime NPF events is the higher fraction of chlorides during nighttime. Chloride is a primarily sourced aerosol component, so it is not influenced by the aerosol formation capacity of the atmosphere at night that reduces the total organic, sulfate, nitrate, and ammonia mass but does not impact chlorides.

Something to highlight is the higher fraction of ammonium compared to nitrates through most of the events. Regions with low NO_x have been previously characterized with higher ammonium fractions compared to nitrates (Du et al., 2015; Liu et al., 2022; Petit et al., 2015; Takami et al., 2005; Topping et al., 2004), while regions with higher NO_x concentrations favour nitrate formation (Hu et al., 2015; Parworth et al., 2015; Poulain et al., 2020; Schlag et al., 2016). The urban vs rural difference in relative mass composition is evident when comparing this study with the aerosol mass fractions observed at an urban site in Sydney (Keywood et al., 2016), in which high nitrate fractions were observed during most of the campaign.